10 TOPS (Dense, INT8) of host-independent AI on one 17mm chip, running inside cameras, drones and robots.
A built-in image processor with HDR and noise reduction keeps detection accurate in low light and glare.
Hardware video encoding at 4K 60 FPS. The device can stream, record and analyze at the same time.
Cameras, displays, Ethernet, USB and sensors wire directly to the chip.
GPUs are the right place to build and test models. SDK qb takes those PyTorch, TensorFlow and ONNX models and compiles them for REGULUS. Over 490 model architectures across CNN, RNN, and transformer have already been validated.
SDK qb converts your model and picks precision layer by layer.
A Yocto Linux SDK with a cross-compile toolchain. It works like any embedded Linux target.
Camera, video and display drivers are ready, with GStreamer and OpenCV included.
Run inspection functions on the production line, including OCR and pattern matching, with no inspection PC.
Wildfire watch, search and rescue, crop spraying. Video is analyzed onboard, not after landing.
Only events go upstream. Raw footage stays on-site.
On-board perception within a mobile power budget.
Vision AI where the enclosure, the bill of materials and the heat budget are already set.
Built for platforms where size, weight and power are fixed.
No. It’s a complete processor that runs Linux and your application without a host.
PyTorch, TensorFlow and ONNX models, compiled with SDK qb. More than 490 architectures have been validated across Mobilint hardware.
Request a REGULUS AI EVK for a full Linux system ready for benchmarking.
